[scudo][standalone] Drain the start of a cache, not the end

[cryptoad]

Summary

When we add a pointer to a cache, we add it at the end. By draining the
end, we were effectively putting back into the central freelist what
was just put in the local cache.

This improves performances, with no impact on memory footprint.

compiler-rt/lib/scudo/standalone/local_cache.h

NOINLINE bool refill(PerClass *C, uptr ClassId) {
     C->Count = B->getCount();
     B->copyToArray(C->Chunks);
     destroyBatch(ClassId, B);
     return true;
   }
 
   NOINLINE void drain(PerClass *C, uptr ClassId) {
     const u32 Count = Min(C->MaxCount / 2, C->Count);
-    const uptr FirstIndexToDrain = C->Count - Count;
-    TransferBatch *B = createBatch(ClassId, C->Chunks[FirstIndexToDrain]);
+    TransferBatch *B = createBatch(ClassId, C->Chunks[0]);
     if (UNLIKELY(!B))
       reportOutOfMemory(
           SizeClassAllocator::getSizeByClassId(SizeClassMap::BatchClassId));
-    B->setFromArray(&C->Chunks[FirstIndexToDrain], Count);
+    B->setFromArray(&C->Chunks[0], Count);
     C->Count -= Count;
+    for (uptr I = 0; I < C->Count; I++)
+      C->Chunks[I] = C->Chunks[I + Count];
     Allocator->pushBatch(ClassId, B);
   }
 };
 
 } // namespace scudo
 
 #endif // SCUDO_LOCAL_CACHE_H_